<![CDATA[Accurate quantification of forest carbon stocks is critical for global climate change mitigation initiatives like REDD+. Traditional forest inventory methods are often labor-intensive, costly, and limited in scale, particularly in complex tropical ecosystems such as the Sumatran rainforest. The integration of advanced remote sensing technologies and artificial intelligence (AI) offers a transformative potential for overcoming these limitations. This study aimed to develop and validate a high-resolution model for individual tree detection and above-ground biomass (AGB) estimation in a Sumatran rainforest by synergizing airborne LiDAR data with machine learning algorithms. High-density LiDAR data was acquired over a 10,000-hectare study area. Concurrently, extensive field inventory data from 150 plots were collected to serve as ground truth. A deep learning model, specifically a Convolutional Neural Network (CNN), was trained to perform individual tree crown delineation (ITCD) from the LiDAR-derived canopy height model. Tree-level metrics were then used as predictors in a Random Forest algorithm to estimate AGB, which was calibrated against field-measured biomass. The CNN model successfully identified individual trees with an accuracy of 92.4%. The subsequent Random Forest model demonstrated high predictive power for AGB estimation, yielding a strong coefficient of determination ( = 0.89) and a low Root Mean Square Error (RMSE) of 25.8 Mg/ha. The approach generated a high-resolution (1-meter) AGB map, revealing detailed spatial variations in carbon stock across the landscape. The fusion of AI and LiDAR data provides a highly efficient methodology for forest inventory and AGB mapping in dense tropical rainforests. This approach significantly enhances our capacity to monitor carbon dynamics, forest conservation and climate policy.]]>
Copyrights © 2025
